Workbench
Find out how much AI-matching Improves Your Reconciliations.
You shouldn't have to take a vendor's word for how well their matching works.
That's why we offer two free ways to see how much of your manual matching Matchimus could take on.
.png)
Two Ways to Get Started
We run it for you
Submit your data. Get a tailored Proof of Value report.
Upload a sample to the Workbench and we’ll run it through Matchimus.
What you get:
-
Your baseline vs. boosted match rate, side by side.
-
The mismatches your current process missed.
-
Hours saved per cycle, ready for your steering committee.
Format
Secure upload · Zero-PII
Output
Proof of Value report
Estimate it yourself
Run BenchRec against your own system.
Run the public BenchRec dataset (ACM ICAIF 2023, free on Kaggle) through your current matching process, then score it with our script.
What you get:
-
A match-rate baseline you can compare with any vendor’s published score.
-
An estimate of what AI matching would add, measured against Matchimus’s BenchRec score.
Format
CSV
Best for
Internal validation
Measured on your data
Security
Institutional-Grade Data Governance
You don't need to submit sensitive identifiers or live account data. Everything runs under a Zero-PII framework - anonymised before it reaches us.
01/
Anonymised before upload
Use our mapping schema to anonymise your data before it leaves your environment.
02/
Encrypted in transit
Every upload is encrypted in transit over HTTPS.
03/
Reviewed by our team, delivered to yours
Your report is prepared by our technical team and returned to your secure Workbench - no email attachments, no shared links.
What you get back
The Data for the Business Case
Your Uplift Report gives your finance director and steering committee three specific outputs:
01
Where your rules break down
The exact transaction types and narrative patterns your current logic can't handle - and how many of them there are.
02
How much of your team's time is manual
A projected reduction in manual touchpoints based on your actual exception volume - expressed as FTE time and cost.
03
How your system handles what's coming next
A score showing how well your matching logic copes with new narrative patterns - without needing rule updates or manual fallback.
Your starting point
Know where you stand before you change anything.
Upload your data and get a clear baseline for your match rate, manual intervention rate, and the gap Operartis closes.
.png)
.png)
What you get back
Test it safely, Without touching production.
Everything runs on historical data in an offline layer. Nothing touches your live production systems until you decide it should.
Stage 2
Move to assisted matching (Human-in-the-Loop).
High-confidence matches go to Matchimus. Lower-confidence ones go to your analysts. Your team's workload drops without losing control.
-
Only the matches that need a human get one.
-
Analysts accept or reject - and you track exactly how much time that saves.
What this eliminates:
The reputational and model risk of AI making unsupervised decisions.
Stage 1
Start in shadow mode.
We run alongside your existing process. Your team keeps working as normal; we just compare results.
-
Matchimus produces match recommendations with confidence scores.
-
Your team compares results against your current engine.
-
You receive an uplift report showing exactly where Matchimus outperforms your current logic.
What this eliminates:
The risk of a bad match affecting live operations or causing payment errors.
Your starting point
A staged rollout on your terms.
Stage 1: Shadow mode
-
Runs on your historical reconciliation files
-
Returns an uplift report - match rate, accuracy and exception breakdown
-
You only move to Stage 2 when the numbers hit agreed thresholds
Stage 2: Assisted matching
-
High-confidence matches go to Matchimus
-
You track acceptance rate, time saved and remaining exceptions
-
Automation only expands once the eligible slice hits threshold
Stage 3: Partial automation
-
Start automating the safest transaction category only - exact matches with supporting rules
-
Everything else stays in assisted mode
-
The automated slice only grows after sustained reliability is confirmed
